Anomaly-aware electric vehicle charging and battery storage management using smart-meter data and a GCN-BiLSTM autoencoder
This research develops a smart meter-based residential energy management system that uses advanced deep learning to detect anomalies in electricity consumption data and optimize power usage in homes with electric vehicles and battery storage. The proposed framework addresses the limitations of traditional smart meters, which usually measure and transmit power consumption but do not actively identify abnormal readings or support intelligent energy scheduling. Smart meter data from 900 households are used to design and evaluate the model. The anomaly detection module is based on an autoencoder that integrates Graph Convolutional Networks and Bidirectional Long Short-Term Memory networks. This structure enables the system to learn both the relational and time-dependent behavior of household energy consumption. Missing values and abnormal readings are detected by comparing original and reconstructed consumption patterns. The corrected smart meter data are then applied to a multi-objective power management strategy that controls EV and battery charging/discharging, reduces peak demand, lowers grid energy consumption, and improves the use of renewable energy sources. The results show that the proposed method achieves high anomaly detection performance and provides better energy cost reduction than conventional approaches. Therefore, the study offers a practical framework for reliable, economical, and sustainable residential energy management.